The output transistors are rated 10W in TO-126 packages, FWIW. Each channel dissipates about 6W in the output stage. Low enough to run on a pair of C cell batteries.
I think there may be a mild instability issue. I noted some variable hum at intermediate volume settings. Playing around, I found that if I touch the wires going from the potentiometers to the board inputs, the hum intensity changes, and if I hold them together, the hum stops, there is a little "pop", and the DC offset jumps. I wrapped the wires in foil and grounded it, which kills the hum. Now I can null the DC offset with my (yes) 20 Ohm trimpot and no noise rotating the volume knobs.
I've found that hum I can be a sign of instability. I need to investigate. I didn't see any signs of misbehavior on the scope when I had the amp on the bench.
I'm not sure why you give me a hard time, Dreamth. I think these projects are pretty good, I put a lot of work into them, and I freely share my designs and what I learn in the process.
If it turns out that I can fix this easily, I will package it as-is. TBH, I like it better than the A3. Maybe I will send it to Stereophile so they can make the A4 their new reference headphone amplifier.
I think there may be a mild instability issue. I noted some variable hum at intermediate volume settings. Playing around, I found that if I touch the wires going from the potentiometers to the board inputs, the hum intensity changes, and if I hold them together, the hum stops, there is a little "pop", and the DC offset jumps. I wrapped the wires in foil and grounded it, which kills the hum. Now I can null the DC offset with my (yes) 20 Ohm trimpot and no noise rotating the volume knobs.
I've found that hum I can be a sign of instability. I need to investigate. I didn't see any signs of misbehavior on the scope when I had the amp on the bench.
I'm not sure why you give me a hard time, Dreamth. I think these projects are pretty good, I put a lot of work into them, and I freely share my designs and what I learn in the process.
If it turns out that I can fix this easily, I will package it as-is. TBH, I like it better than the A3. Maybe I will send it to Stereophile so they can make the A4 their new reference headphone amplifier.
At intermediate levels the input impedance is high yet the SNR is not that high...probably twisted pot wires were good enough or just running them next to a ground plane.It's pretty common what you describe, no worries.
I'm not trying to give you any hard time, it's just that you can't stand any criticism...at all.What others would consider just a casual talk you see as an atack to your integrity...I don't know cat's language, sorry!
You only need to pay big money to any magazine to make your amp their official refference amp unless you have intimate friendship relations with someone important there.You don't need to make the best amp in the universe for that.Reviewing audio components is about money...not pride. .
I'm not trying to give you any hard time, it's just that you can't stand any criticism...at all.What others would consider just a casual talk you see as an atack to your integrity...I don't know cat's language, sorry!
You only need to pay big money to any magazine to make your amp their official refference amp unless you have intimate friendship relations with someone important there.You don't need to make the best amp in the universe for that.Reviewing audio components is about money...not pride. .
Last edited:
I spent more time looking for instability last night but came up empty. I discovered that just shielding the one input pair going to the board is enough to get the amp to settle down. As I said, it's not a gradual thing, but a sharp transition from noise to quiet, with a jump in DC offset. Grounding the foil through a capacitor works, as does holding the foil with one hand while touching the chassis plate with a finger on the other hand. Or even just gripping the foil tightly without touching anything else. So I say to myself, "Why does a tiny amount of capacitance from the input leads have this on/off effect on the circuit noise?" It seems like something is oscillating, but I see no evidence of it.
The relay driver is electrically isolated from the rest of the circuit, but touching the ground pin on the (unconnected) LED output connector with a clip lead induces noise in the output. I tried unplugging the input and output connections to one channel to see if there's a ground loop and it made no difference. It's acting like something is floating that shouldn't be. I hooked a clip lead to ground and tried poking around at various points on the board (obviously, ones that are supposed to be grounded) with no consistent results. At one point, putting my finger on the side of the plastic headphone plug was making noise.
There's a lot of FM broadcast signal in the air here. I'm starting to think it's an external thing, but I'm not sure. The good news is the amplifier is well-behaved with the foil hack and sounds really good. The whole subjective review thing is such a can of worms that I hate to say anything, especially in comparison to the A3/HPA-1, for fear of inciting a riot.
As for me and my supposed sensitivity to criticism, you wrote earlier, "You're wrong. The output is the one forcing the emitters of the diamond's feedback trz through dc feedback." When someone says, "You're wrong," I take that seriously. Is there something to learn there? In this case, no, I disagree with your assessment, and I told you so. I'm happy to receive comments along the line of, "Could it be that...?" But if you use the coded words, "You're wrong," you can expect me to push back until I either understand my error or have disproved the claim. I encourage you to avoid armchair psychology. There are things here you don't know about, and I won't say more.
I am cautiously optimistic that the amp will behave if I put it in a box, so that's the plan. With the connectors, I can easily swap out PC boards later if I decide to do another iteration. For now, I don't feel like binning $150 in parts so I will proceed as planned.
The relay driver is electrically isolated from the rest of the circuit, but touching the ground pin on the (unconnected) LED output connector with a clip lead induces noise in the output. I tried unplugging the input and output connections to one channel to see if there's a ground loop and it made no difference. It's acting like something is floating that shouldn't be. I hooked a clip lead to ground and tried poking around at various points on the board (obviously, ones that are supposed to be grounded) with no consistent results. At one point, putting my finger on the side of the plastic headphone plug was making noise.
There's a lot of FM broadcast signal in the air here. I'm starting to think it's an external thing, but I'm not sure. The good news is the amplifier is well-behaved with the foil hack and sounds really good. The whole subjective review thing is such a can of worms that I hate to say anything, especially in comparison to the A3/HPA-1, for fear of inciting a riot.
As for me and my supposed sensitivity to criticism, you wrote earlier, "You're wrong. The output is the one forcing the emitters of the diamond's feedback trz through dc feedback." When someone says, "You're wrong," I take that seriously. Is there something to learn there? In this case, no, I disagree with your assessment, and I told you so. I'm happy to receive comments along the line of, "Could it be that...?" But if you use the coded words, "You're wrong," you can expect me to push back until I either understand my error or have disproved the claim. I encourage you to avoid armchair psychology. There are things here you don't know about, and I won't say more.
I am cautiously optimistic that the amp will behave if I put it in a box, so that's the plan. With the connectors, I can easily swap out PC boards later if I decide to do another iteration. For now, I don't feel like binning $150 in parts so I will proceed as planned.
The beauty of it is that I don't take any words back and still think you were wrong on that because it is mathematically impossible to have 0V at the feedback nod once you have the feedback resistor divider ended in the same 0V ground point .The output was the one forcing the input to -0.7v as a diamond buffer input at the input trz base with no resistor added has identical voltage as the output at the output trz emitter and the most natural way way to fix it was to make rv1 higher in value giving a wider play range also allowing for smaller currents at both the output and input.Making r10 r11 in series with the same rv1 higher you gave a higher play for the emitter currents and let trz auto adjust their bias at a lower current as the output current being smaller the input of the buffer's current gets also smaller. Essentialy it was the same thing except having rv1higher I though you'd have allowed for a significantly higher play range...
As for the rest of it I think it's just your turn to discover the natural disasters resulting from having too many diamond buffers with their huge impedances at a bipolar EMI loving transistor.Not that it's not manageable just useless complication.
As for the rest of it I think it's just your turn to discover the natural disasters resulting from having too many diamond buffers with their huge impedances at a bipolar EMI loving transistor.Not that it's not manageable just useless complication.
If the bases of the input pair were floating, then the whole first stage buffer would be floating and I agree the input would follow the output. But the input isn't floating. It's connected to ground through a 100K resistor. So, for instance, if the input bases float to 1.0V, then you have 1.0V / 100K Ohms = 10uA current flowing in that resistor, coming backwards out of the diamond buffer input. Assuming everything is perfectly matched and balanced, with a current gain of 250,000, that would correspond to 2.5A of output current. Conversely, you'd need to force 2.5A into the output of the buffer to get 10uA flowing in the input.
Ignoring possible issues with the way I implemented the circuit, here's what's actually happening. The diamond buffer current gains aren't matched. The base-emitter offsets aren't matched. The input pair are settling with a 7uV input offset current, which is producing the 700mV DC input offset across the 100K input resistor. The diamond buffer has a voltage gain close to unity, and is driving the feedback node to 700 mV. Without feedback, there'd be 700mV / 332 Ohms = 2.1mA output current flowing in feedback resistor Rg to ground. This is mirrored to the driver stage output. Multiplied by the 100K driver load resistor (Zo), this gives you 210V DC offset. Open loop gain is like 50dB, so the buffer would clip at the rail with 700mV input. With feedback connected, as the amplifier output voltage rises, current flows through Rf and into Rg until voltage across Rg equals just about 1.0V. In both VFAs and CFAs, the action of feedback is to reduce the voltage between the inverting and non-inverting inputs to near-zero. The feedback current will increase until it reaches 2.1mA, minus something like 21uA which is the CFA error current. This 21uA flows in Zo, giving 2.1V DC at the amplifier output.
This is just an engineering argument, nothing personal. I'm happy to hear your counter-arguments, but I can only rely on what I think I know and if you can't convince me, well, we're in a bind. In many debates, there is room for differences of opinion, but this seems like a matter of black-and-white circuit theory, so it should not be necessary to "agree to disagree."
I will also repeat what I said earlier, that I tried to validate what you said with a simplified SPICE model and the simulation did not exhibit the behavior you predict, not that I expected it to.
I don't agree with your analysis, sorry, but I do appreciate your having the interest to comment.
Ignoring possible issues with the way I implemented the circuit, here's what's actually happening. The diamond buffer current gains aren't matched. The base-emitter offsets aren't matched. The input pair are settling with a 7uV input offset current, which is producing the 700mV DC input offset across the 100K input resistor. The diamond buffer has a voltage gain close to unity, and is driving the feedback node to 700 mV. Without feedback, there'd be 700mV / 332 Ohms = 2.1mA output current flowing in feedback resistor Rg to ground. This is mirrored to the driver stage output. Multiplied by the 100K driver load resistor (Zo), this gives you 210V DC offset. Open loop gain is like 50dB, so the buffer would clip at the rail with 700mV input. With feedback connected, as the amplifier output voltage rises, current flows through Rf and into Rg until voltage across Rg equals just about 1.0V. In both VFAs and CFAs, the action of feedback is to reduce the voltage between the inverting and non-inverting inputs to near-zero. The feedback current will increase until it reaches 2.1mA, minus something like 21uA which is the CFA error current. This 21uA flows in Zo, giving 2.1V DC at the amplifier output.
This is just an engineering argument, nothing personal. I'm happy to hear your counter-arguments, but I can only rely on what I think I know and if you can't convince me, well, we're in a bind. In many debates, there is room for differences of opinion, but this seems like a matter of black-and-white circuit theory, so it should not be necessary to "agree to disagree."
I will also repeat what I said earlier, that I tried to validate what you said with a simplified SPICE model and the simulation did not exhibit the behavior you predict, not that I expected it to.
I don't agree with your analysis, sorry, but I do appreciate your having the interest to comment.
Assuming a 250 000 current gain would not be on the leash q3 and q21 will never be able to deliver 2.5 Amps...while approaching 100mA bc550 won't still have an hfe of 500...
It doesn't matter if the current gains aren't matched, the closed loop gain either current or voltage is the one setting the pace for bias.
So...I like your argumentation just refute the arguments.
It doesn't matter if the current gains aren't matched, the closed loop gain either current or voltage is the one setting the pace for bias.
So...I like your argumentation just refute the arguments.
Attachments
I guess my attempt at the Socratic method failed. FWIW, the loop gain in your circuit is determined by the values of the front-end load network (C1/C3/R28) and the feedback resistor (R1), plus whatever impedance the output buffer presents. I don't know the purpose of C1, but it increases the gain progressively below about 500 Hz. Do you need that extra feedback? And why is C3 so large? It's incredibly heavy compensation. What's your phase margin?
you are right, C3 is 82p missed that and corrected later, now reduced to 68p. C1 and R28 combination helps roll off the output at higher frequency, avoids oscillation from mosfets. C1 also introduces additional gain at lower frequency, not a bad thing to have. picked that trick from one of Apex audio designs.
Last edited:
Ok, that makes sense. I can't run your simulations directly because I don't have all the device models handy. I removed a bunch of stuff from your model before to make it work (don't need the servo and the cap multipliers to check out the loop response), but didn't save my work. Maybe you can just post the plots from your sim. I have enough work to keep me busy here, LOL.
Dunno if anyone is actually interested in this project, but... FWIW, I think it's going to be a real winner, after listening for hours this weekend. I started assembling the amplifier today.
I decided I'm running the front-end bias currents way too hot, so I'm going to lower the bias a LOT. SPICE simulation looks good. This should reduce the input offset current by a similar factor. I plan to run 1.0mA in the first stage, and give the second mirror some current gain to make up for it. Feedback resistors need to be adjusted to bring the loop gain back to target. Hopefully the new configuration won't degrade the sound.
I'm listening to the A3 now and I can't say I have any complaints. But on direct comparison, I still think the A4 is a little better on account of having clearer definition in the bottom end. YMMV.
I decided I'm running the front-end bias currents way too hot, so I'm going to lower the bias a LOT. SPICE simulation looks good. This should reduce the input offset current by a similar factor. I plan to run 1.0mA in the first stage, and give the second mirror some current gain to make up for it. Feedback resistors need to be adjusted to bring the loop gain back to target. Hopefully the new configuration won't degrade the sound.
I'm listening to the A3 now and I can't say I have any complaints. But on direct comparison, I still think the A4 is a little better on account of having clearer definition in the bottom end. YMMV.
I remember some of the Brixton amplifiers were mostly known for their lifetime or later 30 years warranty, but their protection circuitry was equally stuffed or even more complex than the power amp itself...
If you aim at highly priced headphones you need to take protection seriously...and a high power amp dc coupled with no protection sounds like a recipee for disaster... 99.99% of the headphones are light loads for even an op amp chip...but at least an op amp has internal over current protection circuitry .Later Crown introduced something of equal complexity in their PA systems if I recall corectly.
Now you're driving light headphones with 10...70 watts power trz with no overcurrent protection betting on the silicon manufacturing process and ignoring that you might accidentaly short the output , then apply pure dc output into some 4k $ headphones....Stereophile guys won't be that happy seeing that.
That is why , even if I don't get all the dc coupled hype , preffer input, feedback and output capacitor coupling ... it makes life significantly easier and I can bet you won't hear the difference with the right capacitors.You can't complain about capacitor's price or added complexity of a protection circuitry when hiting at 4...5k $ headphones.
Just make a cap virtual ground divider and see what I mean...If Quad , QSC and Neve did it everybody can.
If you aim at highly priced headphones you need to take protection seriously...and a high power amp dc coupled with no protection sounds like a recipee for disaster... 99.99% of the headphones are light loads for even an op amp chip...but at least an op amp has internal over current protection circuitry .Later Crown introduced something of equal complexity in their PA systems if I recall corectly.
Now you're driving light headphones with 10...70 watts power trz with no overcurrent protection betting on the silicon manufacturing process and ignoring that you might accidentaly short the output , then apply pure dc output into some 4k $ headphones....Stereophile guys won't be that happy seeing that.
That is why , even if I don't get all the dc coupled hype , preffer input, feedback and output capacitor coupling ... it makes life significantly easier and I can bet you won't hear the difference with the right capacitors.You can't complain about capacitor's price or added complexity of a protection circuitry when hiting at 4...5k $ headphones.
Just make a cap virtual ground divider and see what I mean...If Quad , QSC and Neve did it everybody can.
I got this running just now. The only thing left to do is a little chassis work.
The goal of this project was to keep the general character of the A3, but improve on it. I think I've succeeded. What that means to other people is a matter of interpretation.
I swapped out the A3 for the A2 while I was putting the new amp together and I have to say, I had forgotten how good the A2 is. Build the A2 if you want a very neutral, matter-of-fact amplifier that sounds superb without being overly romantic or lush (but never thin or harsh). Build the A4 if you clear, well-defined bass, and a touch of euphonic warmth. Build the A3 if you want... well... an A3.
I changed out sixteen resistors last night. The input buffer is now running 1.0mA in the first stage. I gave the second mirror some current gain, to set the front-end cascode bias at 2.2mA. Doing so, I had to adjust the feedback network to bring the loop gain back to the target 40 dB. There is enough current there to drive Cdom to 93 V/uS while keeping the front-end in Class A. DC offset at the input with no volume control connected is around 30mV (lower with the attenuators plugged in). Final stage offset with the servo jumper in place is below 1mV. The trim put is too small to completely null the output DC with the servo disconnected, but I'm not worried about it. There's no hum or undignified behavior. I may do a revised version at some point with a more effective offset trim scheme, but it's probably not even needed. I'm very happy with this amp as-is, so we'll see.
I've uploaded my latest changes to GitHub, so knock yourselves out.
Given the apparent lack of interest here, I think I'm going to call it quits on this thread. I've built six headphone amps in the past eighteen months. I still have the Toroidy transformer, so I may build a monster amp, probably based on the A4 topology. If I do that, I'll start a new thread. Thanks for the encouragement and support; it's been real.
As always, your mileage may vary. Past performance is no guarantee of future returns.
I said I was done with this thread, but I'll answer your question.
I didn't sign a contract to build the HPA-1. Ever since you've gotten involved, it's all about the HPA-1. Enough already. You said I would learn something from building the HPA-1, and I did. I learned that simple, low-feedback circuits sound different. Better in some ways, worse in others. I also learned how easy it is to get carried away by your imagination.
The HPA-1 isn't an unusual circuit. I've found examples of prior art that are very similar. It's an audio truism that simpler circuits with less feedback are more sensitive to component variations. Jam spent a year refining the HPA-1. I spent four weeks from concept to completion on the A4. The A4 mops the floor with the A3. Maybe I will feel differently in a day or two. We could argue forever what that means. I prefer to take notes and move on.
I believe in transparency. I've posted a lot here, and people who've written to me privately know I am happy to give advice and share information. I am put off by people who are cryptic. I understand some folks are building the A2 and/or A3. It would be cool to get some news. I worry that people are running into problems on my account.
I need to listen more to this A4. I was blown away by how good it sounded last night. I need to evaluate it more and see if this impression stands up. All my amplifiers have used the same power supply. JLCPCB sends boards in batches of five. By all accounts, the Super Regulator is superb. I know how you feel about it; that's your opinion.
You can order the Super Regulator boards from the diyaudio store. I can share my Gerbers, but the Talema transformers are out of stock globally. I like my layout better, but performance should be the same. The LDO regulators I use on the amp boards are impossible to get right now. If people are interested in trying the A4, I would consider releasing a revised design with, say, cap multipliers in place of the regulators. I would also get rid of the trimpot as it seems unnecessary now.
Mark, I know you have tried without success to get people to build the HPA-1 clone. I did my share of the heavy lifting. If you want to see this developed into a widely-copied project, and you care about the power supply, how about you design a companion regulator board per the production spec? Then go out and get people to build it.
Thanks, folks, for the thumbs up to my earlier post. I'm glad to get some positive feedback.
I didn't sign a contract to build the HPA-1. Ever since you've gotten involved, it's all about the HPA-1. Enough already. You said I would learn something from building the HPA-1, and I did. I learned that simple, low-feedback circuits sound different. Better in some ways, worse in others. I also learned how easy it is to get carried away by your imagination.
The HPA-1 isn't an unusual circuit. I've found examples of prior art that are very similar. It's an audio truism that simpler circuits with less feedback are more sensitive to component variations. Jam spent a year refining the HPA-1. I spent four weeks from concept to completion on the A4. The A4 mops the floor with the A3. Maybe I will feel differently in a day or two. We could argue forever what that means. I prefer to take notes and move on.
I believe in transparency. I've posted a lot here, and people who've written to me privately know I am happy to give advice and share information. I am put off by people who are cryptic. I understand some folks are building the A2 and/or A3. It would be cool to get some news. I worry that people are running into problems on my account.
I need to listen more to this A4. I was blown away by how good it sounded last night. I need to evaluate it more and see if this impression stands up. All my amplifiers have used the same power supply. JLCPCB sends boards in batches of five. By all accounts, the Super Regulator is superb. I know how you feel about it; that's your opinion.
You can order the Super Regulator boards from the diyaudio store. I can share my Gerbers, but the Talema transformers are out of stock globally. I like my layout better, but performance should be the same. The LDO regulators I use on the amp boards are impossible to get right now. If people are interested in trying the A4, I would consider releasing a revised design with, say, cap multipliers in place of the regulators. I would also get rid of the trimpot as it seems unnecessary now.
Mark, I know you have tried without success to get people to build the HPA-1 clone. I did my share of the heavy lifting. If you want to see this developed into a widely-copied project, and you care about the power supply, how about you design a companion regulator board per the production spec? Then go out and get people to build it.
Thanks, folks, for the thumbs up to my earlier post. I'm glad to get some positive feedback.